Application Of Equipment Fault Sound Detection Based On Blind Source Separation

When the power equipment is running, the situation of high voltage and strong electromagnetic field will impose restrictions on traditional contact-type detection system which based on vibration feature. According to this problem, we proposed a non-contact fault sound detection method which is based on audio features. Blind source separation (BSS) is a new method and new technology in the field of signal processing, it refers to a process that estimates the source signal only by the observed signal which is collected by the sensors, not knowing source signals and hybrid system parameters. Due to the potential application value, BSS technology has obtained great attention of many scholars at home and abroad, which has been developing rapidly over the past few decades. At present, BSS has been widely used in the field of speech enhancement and recognition, biological engineering, digital image processing and information security, etc. In this paper, we use the BSS technology in the problem of multiple audio source interference of equipment fault detection, and then study the application of it.In this paper, we firstly take theoretical analysis about linear model of BSS and nonlinear model of BSS, and then make key research of linear model. Linear model is a more ideal mathematical statistical model, there are some common algorithms based on this model. (1)Fixed point algorithm which based on negative entropy measures the independence of two signals through negative entropy. When the value of negative entropy is bigger, the separation signals tend to be more independent. (2)Natural gradient algorithm which based on information maximization turns the problem of information maximization into solving the problem of negative entropy maximization. (3)Joint approximate diagonalization algorithm uses the fourth order cumulant to build the joint matrix, and then estimate the best separation matrix based on principles of approximate diagonalization. We use three algorithms to separate the mixed sound signals, and then analyze the separation signals by performance index, similarity coefficient and audiometry. The simulation result shows that BSS technology is effective to the separation of multiple audio source in equipment fault detection, and we can acquire different separation performance with different methods.Due to the needs of BSS based on nonlinear model in the practical engineering, we preliminarily study blind source separation algorithm of post-nonlinear and algorithm based on back propagation neural network. But the further experimental research needs to be further carried out, and then apply it to the equipment fault sound detection.